What is the difference of H3?

[SrLiath]

No worries, I'm just curious, what's the main difference between H3 and other libs like FastAPI or Express?
It looks tiny, is it faster, or is there something else I'm missing?

[bmhtech07]

It's faster than FastAPI (Node vs Python).

It's lighter weight (and less opinionated) than express.

[SrLiath]

cool, i will try do somes test, thnks

ps i said it wrong, I meant fastify (but cool to know) ;P

[SrLiath]

for future purposes, in case anyone comes here, my tests returned:

Express - Req/sec: 1892.67
Express - Max latency: 4421 ms
Express - Errors: 88085
version: 5.1.0
package size: 52.1 kB
unpacked size: 197.0 kB

Fastify - Req/sec: 37212.54
Fastify - Max latency: 614 ms
Fastify - Errors: 0
version: 5.3.3
package size: 458.6 kB
unpacked size: 2.6 MB

H3 - Req/sec: 24247.34
H3 - Max latency: 628 ms
H3 - Errors: 0
version: 1.15.3
package size: 121.5 kB
unpacked size: 527.6 kB

(I ran the same test about 10 times, and the result was consistent. It’s a good option to use, I'm really impressed, even though I've tested simple routes, it's very light and has considerable performance, I used Autocannon to make the test.)

test code (all libs was the basic of your documentation returning a "ok"):

import autocannon from 'autocannon'
const targets = [
  { name: 'Express', url: 'http://server:3001' },
  { name: 'Fastify', url: 'http://server:3002' },
  { name: 'H3', url: 'http://server:3003' },
]

for (const { name, url } of targets) {
  await new Promise((resolve) => {
    autocannon(
      {
        url,
        connections: 500,
        duration: 30,
        pipelining: 10,
        timeout: 1,
      },
      (err, result) => {
        if (err) console.error(err)
        console.log(`${name} - Req/sec:`, result.requests.average)
        console.log(`${name} - Max latency:`, result.latency.max, 'ms')
        console.log(`${name} - Errors:`, result.errors)
        resolve()
      }
    )
  })
}

[gulshan]

Can you try the H3 v2? Like this in the dependency-

    "h3": "npm:h3-nightly@2x",

[SrLiath]

Is there a serverless version of h3? It would be interesting

[SrLiath]

I didn't even know this size was possible lol

package size: 121.5 kB
unpacked size: 527.6 kB

My tests still returned something significant, despite being as light as the previous one and still returning an extremely better latency compared to the express, this one was more mixed, the average it returned to me was:
H3 - Req/sec: 30398.4
H3 - Max latency: 460 ms
H3 - Errors: 0

It is better than Fastify and but is not more stable than Express. I ran a test comparing Express and H3 v2 with 1,500 simultaneous connections using Autocannon for 30 seconds(something in 45000 connections). In the worst-case scenario, H3 returned 198,973 errors, while Express returned 114,300 errors. I’ll wait for some new releases, but it’s clear that I’ll try using H3 to see if it performs the same way in a more complex system.

[SrLiath]

I was a bit confused, so I decided to increase the test duration and ran it continuously for 2 seconds with 500 simultaneous connections to see if the issues would repeat. In the worst-case scenario, H3 handled around 24,000 requests per second but showed a maximum latency of about 2,845 milliseconds and returned over 14,000 errors. Express managed only about 2,900 requests per second, had a lower maximum latency around 1,484 milliseconds, but suffered from a high number of errors, more than 65,000. Fastify outperformed both, handling around 44,600 requests per second with a maximum latency below 500 milliseconds and no errors at all.
H3 performs better than Express in terms of throughput but still experiences a significant number of errors and high latency, which means it’s not as stable under pressure. I plan to keep testing, especially in more complex scenarios

[pi0]

i suggest to wait and test against upcoming h3 v2 or srvx directly. h3 v2 will use srvx as sever engine and much smaller/faster.